U.S. patent number 8,606,727 [Application Number 12/942,783] was granted by the patent office on 2013-12-10 for method and system for traffic prediction based on space-time relation.
This patent grant is currently assigned to NEC (China) Co., Ltd.. The grantee listed for this patent is Junjian He, Weisong Hu, Xiaowei Liu, Jia Rao, Shaoya Wang, Tao Wu. Invention is credited to Junjian He, Weisong Hu, Xiaowei Liu, Jia Rao, Shaoya Wang, Tao Wu.
United States Patent |
8,606,727 |
Wu , et al. |
December 10, 2013 |
**Please see images for:
( Certificate of Correction ) ** |
Method and system for traffic prediction based on space-time
relation
Abstract
A system and method for traffic prediction based on space-time
relation are disclosed. The system comprises a section spatial
influence determining section for determining, for each of a
plurality of sections to be predicted, spatial influences on the
section by its neighboring sections; a traffic prediction model
establishment section for establishing, for each of the plurality
of sections to be predicted, a traffic prediction model by using
the determined spatial influences and historical traffic data of
the plurality of sections; and a traffic prediction section for
predicting traffic of each of the plurality of sections to be
predicted for a future time period by using real-time traffic data
and the traffic prediction model. An apparatus and method for
determining spatial influences among sections, as well as an
apparatus and method for traffic prediction, are also disclosed.
With the present invention, a spatial influence of a section can be
used as a spatial operator and a time sequence model can be
incorporated, such that the influences on a current section by its
neighboring section for a plurality of spatial orders can be taken
into account. In this way, the traffic condition in a spatial scope
can be measured more practically, so as to improve accuracy of
prediction.
Inventors: |
Wu; Tao (Beijing,
CN), Wang; Shaoya (Beijing, CN), He;
Junjian (Beijing, CN), Hu; Weisong (Beijing,
CN), Rao; Jia (Beijing, CN), Liu;
Xiaowei (Beijing, CN) |
Applicant: |
Name |
City |
State |
Country |
Type |
Wu; Tao
Wang; Shaoya
He; Junjian
Hu; Weisong
Rao; Jia
Liu; Xiaowei |
Beijing
Beijing
Beijing
Beijing
Beijing
Beijing |
N/A
N/A
N/A
N/A
N/A
N/A |
CN
CN
CN
CN
CN
CN |
|
|
Assignee: |
NEC (China) Co., Ltd. (Beijing,
CN)
|
Family
ID: |
44174507 |
Appl.
No.: |
12/942,783 |
Filed: |
November 9, 2010 |
Prior Publication Data
|
|
|
|
Document
Identifier |
Publication Date |
|
US 20110161261 A1 |
Jun 30, 2011 |
|
Foreign Application Priority Data
|
|
|
|
|
Dec 28, 2009 [CN] |
|
|
2009 1 0265617 |
|
Current U.S.
Class: |
706/12 |
Current CPC
Class: |
G08G
1/0104 (20130101) |
Current International
Class: |
G06F
15/18 (20060101) |
References Cited
[Referenced By]
U.S. Patent Documents
Foreign Patent Documents
|
|
|
|
|
|
|
11-316891 |
|
Nov 1999 |
|
JP |
|
2000-285362 |
|
Oct 2000 |
|
JP |
|
2005-227972 |
|
Aug 2005 |
|
JP |
|
2006-214974 |
|
Aug 2006 |
|
JP |
|
2008-123474 |
|
May 2008 |
|
JP |
|
2010-072986 |
|
Apr 2010 |
|
JP |
|
Other References
Kamarianakis, Prastacos, "Space-time Modeling of Traffic Flow",
Computers and Geosciences, vol. 31, 2005, pp. 119-133. cited by
examiner .
Lin, Huang, Zhu, Wang, "The Application of Space-Time ARIMA Model
on Traffic Flow Forecasting", IEEE, Proceedings of the Eighth
International Conference on Machine Learning and Cybernetics,
Baoding, Jul. 2009, pp. 3408-3412. cited by examiner .
Nguyen, Scherer, "Imputation Techniques to Account for Missing Data
in Support of Intelligent Transporation Systems Applications",
Research Report No. UVACTS-13-0-78, University of Virginia,
Charlottesville, VA, 2003, pp. 1-114. cited by examiner .
Berg, Hegyi, Schutter, Hellendoorn, "Integrated traffic control for
mixed urban and freeway networks: A model predictivce control
approach", European Journal of Transport and Infrastructure
Research, vol. 7, No. 3, 2007, pp. 223-250. cited by examiner .
Office Action issued Jun. 21, 2012, by the People Republic of China
in counterpart Chinese Application No. 2010-247289. cited by
applicant .
Shiliang Sun, et al., "Traffic Flow Forecasting Using a
Spatio-temporal Bayesian Network Predictor", Proceedings of ICANN
2005, pp. 273-278. cited by applicant .
Yiannis I. Kamarianakis, et al., "Space-Time Modeling of Traffic
Flow", IEEE Transactions on Fuzzy Systems, 2002, pp. 1-22. cited by
applicant.
|
Primary Examiner: Chaki; Kakali
Assistant Examiner: Hanchak; Walter
Attorney, Agent or Firm: Sughrue Mion, PLLC
Claims
What is claimed is:
1. A method for determining spatial influences among sections,
comprising: determining, for each of sections in a road network, a
spatial scope having influence on the section, wherein the spatial
scope is of a spatial order of N, which is an integer equal to or
greater than 1; extracting, from the road network, neighboring
sections of the section within the determined spatial scope, as
N-th order influential sections for the section; classifying the
spatial relation between each of the sections and each of its N-th
order influential sections into one of predefined types of spatial
relation; performing, for the classified type of spatial relation,
correlation analysis based on historical traffic data of the
section and its N-th order influential sections of this type of
spatial relation, to learn a correlation between the section and
its N-th order influential sections for this type of spatial
relation; and step of determining spatial influences of spatial
order N for the section based on the learned correlation, wherein
each of the spatial influences reflects an extent to which the
section is influenced by one of its N-th order influential
sections.
2. The method of claim 1, wherein in the spatial scope determining
operation, the spatial scope having influence on the section is
determined according to the relative spatial locations of the
sections in the road network.
3. The method of claim 1, wherein in the spatial scope determining
operation, for each of the sections in the road network, a spatial
scope that can be reached within a preset time period by starting
travel from the section is determined as the spatial scope having
influence on the section.
4. The method of claim 1, wherein the predefined types of spatial
relation comprise no relation, precede straight, precede merge,
precede intersect, precede diverge, succeed straight, succeed
merge, succeed intersect, and succeed diverge; or the predefined
types of spatial relation comprise straightforward, left turn and
right turn.
5. The method of claim 1, wherein in the spatial influence
determining operation, each of the N-th order influential sections
of a section is allocated with an influential weight based on the
correlation between the section and the N-th order influential
section, and the spatial influence on the section by the N-th order
influential section is determined using the influential weight.
6. The method of claim 1, wherein the spatial influences on a
section by its N-th order influential sections are represented in a
vector having a dimension equal to the number of its N-th order
influential sections.
7. The method of claim 1, wherein the spatial influences among all
of a plurality of sections are represented in a M.times.M matrix, M
being equal to the number of the plurality of sections and each row
or each column of the matrix representing the spatial influences on
one of the plurality of sections by its N-th order influential
sections.
8. The method of claim 1, wherein, for a changed spatial order N,
spatial influences for the changed spatial order N are determined
for each of the sections through the spatial scope determining
operation, the influential section extraction operation, the
spatial relation determining operation, the correlation learning
operation and the spatial influence determining operation.
9. The method of claim 1, further comprising: storing, for each of
the sections, the determined spatial influences for at least one
spatial order N.
10. The method of claim 1, wherein the historical traffic data
comprise, for a particular time period in a day, at least one of
the following historical traffic data for each section: a travel
speed at which a vehicle travels along the section, a travel time
period required for a vehicle to travel through the section, a
section congestion indication representing a ratio between an
actual travel time period required by a vehicle to actually travel
through the section and a free flow travel time period expected for
a vehicle to travel through the section in a free flow condition,
or representing a ratio between an actual travel speed at which a
vehicle actually travels along the section and a free flow travel
speed at which a vehicle travels along the section in a free flow
condition.
11. The method of claim 1, wherein a section comprises one of: a
link as basic road element of a road network, a road segment
obtained by analyzing a road network and building a mapping between
road segments and links; and a road segment from one intersection
to another adjacent intersection in the road network.
12. A traffic prediction method, comprising: obtaining real-time
traffic data for a plurality of sections within one or more time
periods, as prediction input; selecting a traffic prediction model
for each of the sections whose traffic is to be predicted, based on
a future time period for which the prediction is to be made and/or
a specified time order and/or spatial order, wherein the traffic
prediction model is a time sequence model considering spatial
relation, and the spatial relation is represented by the spatial
influences among the sections as determined by a method for
determining spatial influences among sections according to claim 1;
and predicting traffic of each of the sections for a future time
period after a specified time period by using the prediction input
and the selected traffic prediction model.
13. The method of claim 12, wherein the traffic prediction model
comprises a Space-Time Auto Regression (STAR) model or a Space-Time
Auto Regression Moving Average (STARMA) model.
14. The method of claim 12, further comprising, after the
prediction input obtaining operation: analyzing the difference
between the obtained real-time traffic data and the historical
traffic data, adjusting the obtained real-time traffic data based
on the analysis result, and using the adjusted real-time traffic
data as the prediction input.
15. The method of claim 14, wherein in the data difference analysis
operation, the obtained real-time traffic data is adjusted by way
of statistical averaging.
16. An apparatus for determining spatial influences among sections,
comprising: a spatial scope determining unit, implemented by a
processor, configured to determine, for each of sections in a road
network, a spatial scope having influence on the section, wherein
the spatial scope is of a spatial order of N, which is an integer
equal to or greater than 1; an influential section extraction unit
configured to extract, from the road network, neighboring sections
of the section within the determined spatial scope, as N-th order
influential sections for the section; a spatial relation
determining unit configured to classify the spatial relation
between each of the sections and its N-th order influential
sections into one of predefined types of spatial relation; a
correlation learning unit configured to perform, for the classified
type of spatial relation, correlation analysis on historical
traffic data of the section and its N-th order influential sections
of this type of spatial relation, to learn a correlation between
the section and its N-th order influential sections for this type
of spatial relation; and a spatial influence determining unit
configured to determine spatial influences for the N-th order
influential sections of the section based on the learned
correlation, wherein each of the spatial influences reflects an
extent to which the section is influenced by one of its N-th order
influential sections.
17. The apparatus of claim 16, wherein the spatial scope
determining unit determines the spatial scope having influence on
the section according to the relative spatial locations of the
sections in the road network.
18. The apparatus of claim 16, wherein the spatial scope
determining unit determines, for each of the sections in the road
network, a spatial scope that can be reached within a preset time
period by starting travel from the section as the spatial scope
having influence on the section.
19. The apparatus of claim 16, wherein the spatial influence
determining unit allocates to each of the N-th order influential
sections of a section with an influential weight based on the
correlation between the section and the N-th order influential
section, and determines the spatial influence on the section by its
N-th order influential section using the influential weights.
20. A traffic prediction apparatus, comprising: a prediction input
obtaining unit configured to obtain real-time traffic data for a
plurality of sections within one or more time periods, as
prediction input; a traffic prediction model selection unit
configured to select a traffic prediction model for each of the
sections whose traffic is to be predicted, based on a future time
period for which the prediction is to be made and/or a specified
time order and/or spatial order, wherein the traffic prediction
model is a time sequence model considering spatial relation, and
the spatial relation is represented by spatial influences among the
sections as determined by an apparatus for determining spatial
influences among sections according to claim 16; and a traffic
prediction unit configured to predict traffic of each of the
section for a future time period after a specified time period by
using the prediction input and the selected traffic prediction
model.
21. The apparatus of claim 20, further comprising: a data
difference analysis unit configured to analyze the difference
between the obtained real-time traffic data and the historical
traffic data, adjusting the obtained real-time traffic data based
on the analysis result, and using the adjusted real-time traffic
data as the prediction input.
22. A method for traffic prediction based on space-time relation,
comprising: determining, for each of a plurality of sections to be
predicted, spatial influences on the section by its neighboring
sections, by a method for determining spatial influences among
sections according to claim 1; establishing, for each of the
plurality of sections to be predicted, a traffic prediction model
by using the spatial influences determined at the section spatial
influence determining operation and historical traffic data of the
plurality of sections; and predicting traffic of each of the
plurality of sections to be predicted for a future time period by
using real-time traffic data and the traffic prediction model
established at the traffic prediction model establishment
operation.
23. A system for traffic prediction based on space-time relation,
comprising: a section spatial influence determining section,
implemented by a processor, configured to determine, for each of a
plurality of sections to be predicted, spatial influences on the
section by its neighboring sections, by an apparatus for
determining spatial influences among sections according to claim
16; a traffic prediction model establishment section configured to
establish, for each of the plurality of sections to be predicted, a
traffic prediction model by using the spatial influences determined
at the section spatial influence determining section and historical
traffic data of the plurality of sections; and a traffic prediction
section configured to predict traffic of each of the plurality of
sections to be predicted for a future time period by using
real-time traffic data and the traffic prediction model established
by the traffic prediction model establishment section.
24. The method of claim 1, wherein the spatial relation between
each of the sections and each of its N-th order influential
sections are classified into different groups based on the spatial
relation.
25. The apparatus of claim 16, wherein the spatial relation between
each of the sections and each of its N-th order influential
sections are classified into different groups based on the spatial
relation.
Description
FIELD OF THE INVENTION
The invention relates to traffic information prediction, and more
particularly, to a technology for predicting traffic information
based on space-time relation.
BACKGROUND OF THE INVENTION
In modern society, automobiles are becoming increasingly widespread
with the rapid economic growth, which imposes heavy pressures on
urban traffic and causes severe traffic jams. It is an urgent issue
to mitigate traffic congestions, so as to reduce travel time for
automobile drivers, reduce fuel consumption, improve economic
efficiency of a city and facilitate environment protection. Thus,
the traffic information service system plays an important role in
urban intelligent transport system. Prediction of traffic
information is a core functionality of the traffic information
service system, which is intended to mine history patterns of
traffic information, predict urban traffic condition in near future
and compensate delays in a traffic information service system. It
also enables the drivers to be aware of the future traffic
condition and drive in a stable mood. Furthermore, it is of
significance that the prediction is based on the real-time traffic
information gathering system while extending the real-time traffic
information service to both the past and the future.
Currently, the rapid development of mobile communication technology
and the popularization of GPS technology provide potentials for
accurately gathering real-time traffic. In general, such a
technology can be classified into a fixed probing technology and a
mobile probing technology. The fixed probing technology involves
gathering real-time traffic information and monitoring traffic
conditions by fixed equipments, such as loops, RTMS (Remote Traffic
Microwave Sensor) and monitoring cameras. On the other hand, the
mobile probing technology comprises probe vehicle technology and
probe mobile terminal technology. A probe vehicle refers to a
vehicle equipped with both a GPS module and a mobile communication
module; and the probe vehicle technology involves obtaining in
real-time vehicle-related data such as geographical location data
of a probe vehicle, uploading the data to a data center regularly
via the mobile communication network, performing a map matching,
path finding and traffic information fusion at a server side and,
finally, disseminate real-time traffic information to the user
terminals. In contrast, the mobile terminal probing technology
involves obtaining cell locations of a large amount of mobile
terminal users by means of base station positioning in a mobile
communication network, analyzing users' behavior patterns, finding
out a sequence of position points which can reflect the traffic
condition, calculating real-time traffic information with reference
to digital map data and providing real-time traffic information
service.
However, the existing technologies for acquiring real-time traffic
information cannot satisfy all user requirements. In most cases, a
driver desires to know not only the current traffic condition, but
also the traffic condition in the near future, so as to avoid
congested roads. In addition, the current technologies for
acquiring real-time traffic information suffer from a certain
period of delay due to time consumptions during data transmission
and system calculation, while the real-time traffic condition may
vary rapidly. Therefore, the prediction of traffic information
becomes particularly important in practical applications and thus
becomes in recent years a topic of interest in world-wide research
for intelligent transport systems.
Generally, the traffic information prediction technology
establishes a suitable prediction model, such as a time sequence
model, a neural network model, a Bayesian model, a fuzzy
mathematical model, based on accumulated historical traffic
information, so as to perform information prediction. A practical,
applicable traffic information prediction system should satisfy two
aspects of functionalities. First, from the perspective of time
length of prediction, it is necessary to support short-term,
mid-term and long-term predictions. Second, from the perspective of
spatial scope of prediction, it is necessary to support traffic
information prediction for the entire road network, rather than
merely for arterial roads or highways. Meanwhile, the road network
is complicated and has a large amount of data; and a prediction
model itself is highly complicated. Thus, it is a vital but
difficult research topic to achieve traffic information prediction
with high performance and high accuracy.
There have been some patents and papers involving methods and
models for traffic information prediction. Most of these methods,
however, are based on arterial roads or highways, not complete road
network, a low applicability and a relatively low model complexity.
Further, most of these methods did not consider the spatial
relations in a road network, but rather perform prediction modeling
on each of individual roads by time series analysis, fuzzy
mathematics, etc. For a few space-time prediction researches, there
are also a variety of drawbacks. The related patents and papers
will be introduced in the following.
Patent Document 1, "Travel-time Prediction Apparatus, Travel-time
Prediction Method, Traffic Information Providing System and
Program", US Patent No. 20080097686(A1), discloses a method for
traffic information prediction based on an Auto-Regression (AR)
time series model. It utilizes a single link as a processing
element, establishes a time series sample data for link travel time
based on historical traffic condition data and set up an AR model
for traffic information prediction.
Patent Document 2, "System and Method of Predicting Traffic Speed
Based on Speed of Neighboring Link", US Patent No. 20080033630(A1),
discloses a method for predicting current link condition based on
conditions of neighboring links. According to this solution,
adjacent links at two endpoints of a single current link are
calculated in advance, and then a relation between the current
traveling speed on the current link and the traveling speeds on the
adjacent links is derived from previous traveling speed on each of
the links. Finally, the traffic condition prediction can be
performed based on such a model.
Non-patent Document 1, "Traffic Flow Forecasting Using a
Spatio-temporal Bayesian Network Predictor", Proceeding of ICANN
2005, discloses a method for traffic information prediction based
on a space-time Bayesian network.
Non-patent Document 2, "Space Time Modeling of Traffic Flow", IEEE
TRANSACTIONS ON FUZZY SYSTEMS, 2002, discloses a method for
space-time modeling of traffic flow. According to this method,
spatial features are incorporated into a prediction model by using
a matrix of weights based on distance estimation, and then a
space-time auto-regression moving average model can be established
for short-term prediction of traffic information.
Among the above prior art solutions, Patent Document 1 establishes,
on a link basis, an auto-regression model for travel time on each
link for prediction. This solution, however, only considers the
time domain while completely ignoring the interrelation among the
road links. Moreover, it only reflects the historical traffic
characteristics of a single link while failing to represent the
influence of changes in traffic conditions of neighboring links on
the current link. Patent Document 2 calculates the traveling speed
on the current link based on the traveling speed on adjacent links.
In fact, this solution involves no prediction of future traffic
condition, but only calculation of traveling speeds on neighboring
links based on a known link traveling speed. However, for traffic
condition, the same traveling speed may imply different levels of
congestion. It is thus improper to utilize speed as a sample value.
Non-patent Document 1 uses a Bayesian network which is complicated
in structure and very inefficient when applied to traffic
information prediction for a large scale road network. Non-patent
Document 2 teaches to distinguish the levels of influences of
spatial relation based on distance while ignoring the influence of
a key connection node on the road traffic flow. Meanwhile, this
solution measures the road condition with traffic flow only,
without taking into account that different levels of roads have
themselves different capacities for accommodating traffic
flows.
To summarize, the existing solutions are inadequate for traffic
prediction, particularly for mining spatial relations, including
determining the scope of spatial influence, allocating weights for
spatial influence objects, unifying criteria for evaluation of
traffic condition, as well as mining the relation between the
historical traffic conditions of the current road and the roads
within the scope of spatial influence. Further, some of the
solutions select prediction models which are not extendable, so
that the system efficiency decreases exponentially with the
increase in prediction scope.
Obviously, it is insufficient to only establish a traffic
information prediction model based on historical data and perform
time sequence analysis on a single segment. The influences of
precede/succeed roads should be considered as there are strong
mutual influences among the road segments in the road network. For
example, a road will be very likely to be congested if its succeed
road is congested, and will be very likely to be unblocked if its
precede road is not congested. Thus, it is desired to establish a
traffic information prediction model taking into account analysis
models in both space and time domains.
The time sequence model is a common prediction and control model,
which finds out statistical regularities for prediction based on
historical data. A Space-Time Auto Regression Moving Average
(STARMA) model is a general time sequence model considering spatial
relation, which is suitable for analysis space-time statistical
data. This model is applicable in various fields such as regional
economics and weather forecasting analysis. A core issue in
utilization of this model is how to define the spatial relation,
including which object to be used in spatial analysis, how to
determine a spatial scope which has influence on a spatial object,
and how to determine influence weights for individual spatial
objects in the scope.
The present invention is directed to a method for traffic
prediction based on space-time relation with high performance and
high accuracy, which takes fully into account spatial
characteristics of a road traffic network
SUMMARY OF THE INVENTION
To solve the above problems, according to an aspect of the present
invention, a method for determining spatial influences among
sections is provided, which comprise: spatial scope determining
step of determining, for each of sections in a road network, a
spatial scope having influence on the section, wherein the spatial
scope is of a spatial order of N, which is an integer equal to or
greater than 1; an influential section extraction step of
extracting, from the road network, neighboring sections of the
section within the determined spatial scope, as N-th order
influential sections for the section; a spatial relation
determining step of classifying the spatial relation between each
of the sections and each of its N-th order influential sections
into one of predefined types of spatial relation; a correlation
learning step of performing, for the classified type of spatial
relation, correlation analysis on historical traffic data of the
section and its N-th order influential sections of this type of
spatial relation, to learn a section correlation between the
section and its N-th order influential sections for this type of
spatial relation; and a spatial influence determining step of
determining spatial influences of spatial order N for the section
based on the learned section correlation, wherein each of the
spatial influences reflects an extent to which the section is
influenced by one of its N-th order influential sections.
In this way, the spatial relation within the road traffic network
itself is fully utilized and the influence of changes in traffic
conditions of neighboring sections on the current section is
considered for spatial scopes of various spatial orders. As such,
in actual prediction, any change in traffic condition at a node or
on a section can be rapidly reflected in the corresponding spatial
scope, which is impossible for the prediction algorithm considering
only one single section.
In an embodiment, in the spatial scope determining step, the
spatial scope having influence on the section is determined
according to the relative spatial locations among the sections in
the road network.
In this way, the influence scope can be determined from the
perspective of relative spatial locations of segments in the road
network, e.g., considering such factors as direct adjacency and/or
relative distance from each other.
In an embodiment, in the spatial scope determining step, for each
of the sections in the road network, a spatial scope reachable from
the section within a preset time period is determined as the
spatial scope having influence on the section.
In this way, the influence scope can be determined in terms of
time. For example, a spatial scope can be determined as reachable
within a preset time period by starting traveling from the current
section at the current speed or an average speed based on
historical data. The preset time period can be for example a
traffic information gathering period or a multiple thereof, to
further facilitate analysis of traffic data.
In an embodiment, in the correlation learning step, the spatial
relation between each of the sections and each of its N-th order
influential sections is classified into one of predefined types of
spatial relation and, for the classified type of spatial relation,
correlation analysis is performed on historical traffic data of the
section and its N-th order influential sections of this type of
spatial relation to learn a section correlation between the section
and its N-th order influential sections for this type of spatial
relation. In an embodiment, the predefined types of spatial
relation comprise no relation, precede straight, precede merge,
precede intersect, precede diverge, succeed straight, succeed
merge, succeed intersect, and succeed diverge;
Alternatively, the predefined types of spatial relation comprise
straightforward, left turn and right turn.
In this way, the spatial relations between a section and its
influential sections can be classified into a variety of types,
such that different influences resulted from different spatial
relations can be considered differentially.
In an embodiment, in the spatial influence determining step, each
of the N-th order influential sections of a section is allocated
with an influential weight based on the correlation between the
section and the N-th order influential section, and the spatial
influence on the section by the N-th order influential section is
determined based on the influential weight.
In this way, the extent to which the current section is influenced
by each of its influential sections can be reflected with respect
to different spatial relations.
In an embodiment, the spatial influences on a section by its N-th
order influential sections are represented in a vector having a
dimension equal to the number of its N-th order influential
sections. Alternatively, the spatial influences among all of a
plurality of sections are represented in a M.times.M matrix, M
being equal to the number of the plurality of sections and each row
or each column of the matrix representing the spatial influences on
one of the plurality of sections by its N-th order influential
sections.
In this way, the spatial relations among a plurality of sections
can be reflected intuitively and compactly in a vector or a matrix,
which can be conveniently substituted as a spatial operator into
the time sequence model, so as to simplify the subsequent processes
of modeling and prediction.
In an embodiment, the above method further determines, for a
changed spatial order N, spatial influences for the changed spatial
order N for each of the sections by the spatial scope determining
step, the influential section extraction step, the spatial relation
determining step, the correlation learning step and the spatial
influence determining step. The above method further comprises a
storage step of storing, for each of the sections, the determined
spatial influences for at least one spatial order N.
In this way, the spatial relations among all the sections in the
road traffic network can be fully considered to obtain, for each of
a plurality of different spatial orders, influence of the changes
in traffic condition of neighboring sections on the current
section, such that the overall traffic condition can be reflected.
Additionally, in actual prediction, it is possible to select a
suitable spatial order based on the time period or traffic
condition to be predicted, so as to determine a spatial scope to be
considered for prediction. As such, the traffic prediction can be
more flexible and effective.
In an embodiment, the historical traffic data comprise, for a
particular time period in a day, at least one of the following
historical traffic data for each section: a travel speed at which a
vehicle travels along the section, a travel time period a vehicle
requires for passing through the section, a section congestion
indication representing a ratio between an actual travel time
period a vehicle requires for passing through the section and a
free flow travel time period a vehicle requires for passing through
the section in a free flow condition, or representing a ratio
between an actual travel speed at which a vehicle actually travels
along the section and a free flow travel speed at which a vehicle
travels along the section in a free flow condition.
As for traffic condition, the same travel speed/travel time may
indicate different congestion levels. For example, the rated speeds
for arterial roads and side roads are inherently and dramatically
different from each other. Thus, the congestion level of a road
cannot be properly reflected by only using speed/travel time as
sample. According to the present invention, the congestion
indication of a road is used as historical traffic data for
analysis. In this way, the criterion for measuring traffic in a
spatial scope is unified, and the traffic in the spatial scope can
be measured more accurately, leading to an improved accuracy of
prediction.
In an embodiment, a section comprises one of: a link as basic road
element of a road network, a road segment obtained by analyzing a
road network and building a mapping between road segments and
links; and a road section from one intersection to another adjacent
intersection in the road network.
In this way, the present invention employs a road section between
road nodes, such as intersections which are considered as more
important in the real world, as a basic data object, rather than
based on the conventional link which has shorter length and less
stable traffic characteristics. In addition, by using a road
section obtained by restructuring links, it is possible to utilize
a reduced number of integrated road sections as basic data objects,
so as to improve calculation efficiency as well as prediction
accuracy.
It is thus possible to establish, for each section, prediction
models for different time scopes and spatial scopes by considering
different situations of the section in different time periods,
leading to a more flexible and effective traffic prediction.
According to another aspect of the present invention, a traffic
prediction method is is provided, which comprises: a prediction
input obtaining step of obtaining real-time traffic data for a
plurality of sections within one or more time periods as a
prediction input; a traffic prediction model selection step of
selecting a traffic prediction model for each of the sections whose
traffic is to be predicted, based on a future time period for which
the prediction is to be made and/or a specified time order and/or
spatial order, wherein the traffic prediction model is a time
sequence model incorporating spatial relation, and the spatial
relation is represented by the spatial influences among the
sections as determined by the above method for determining spatial
influences among segments; and a traffic prediction step of
predicting traffic of each of the sections for a future time period
after a specified time period by using the prediction input and the
selected traffic prediction model.
The prediction may be more flexible by selecting the prediction
model based on a future time period for which the prediction is to
be made and/or a specified time order and/or spatial order.
In an embodiment, the traffic prediction model comprises a
Space-Time Auto Regression (STAR) model or a Space-Time Auto
Regression Moving Average (STARMA) model.
Herein, the STAR and the STARMA models are both general time
sequence models considering spatial relation, which are suitable
for analyzing space-time statistical data and mining statistical
patterns for prediction based on historical data. The present
invention employs such general time sequence models incorporating
spatial relation, capable of introducing a novel spatial operator,
without modifying basic models, to reflect the influence of changes
in neighboring traffic on the current section. Accordingly, the
accuracy of prediction can be improved.
In an embodiment, the method further comprises, after the real-time
traffic data obtaining step, a data difference analysis step of
analyzing the difference between the obtained real-time traffic
data and the historical traffic data, adjusting the obtained
real-time traffic data based on the analysis result, and using the
adjusted real-time traffic data as the prediction input. Herein,
the real-time traffic data is adjusted by means of statistical
averaging.
In this way, it is possible to preclude improper or erroneous data
from the real-time traffic data, such that the accuracy of the
prediction input and thus the accuracy of the prediction result can
be improved.
An apparatus for determining spatial influences among sections and
an apparatus for traffic prediction are also provided.
In addition, the present invention discloses a method and system
for traffic prediction.
To summarize, the present invention has the following advantages:
The performance can be greatly improved by, based on a reduced
number of integrated sections, using spatial influence of section
as a spatial operator and by employing a specific time sequence
model, such as the STARMA model. The spatial relations of multiple
orders are incorporated. In actual prediction, any change in
traffic condition at a node or on a section can be rapidly
reflected in the corresponding spatial scope, which is impossible
for the prediction algorithm considering only one single section.
The influences of neighboring sections of multiple spatial orders
on the current section can be considered, which can be applied in
prediction of future traffic and in compensation of calculation for
current traffic to increase traffic coverage. The concept of
congestion indication is introduced to measure traffic condition in
a spatial scope even more practically, thereby improving the
prediction accuracy. The system is designed for the entire road
network and is thus highly applicable.
BRIEF DESCRIPTION OF THE DRAWINGS
The above and further objects, features and advantages can be more
apparent from the following description of the preferred
embodiments with reference to the figures, in which:
FIG. 1 is a diagram showing the configuration of a traffic
prediction system;
FIG. 2 is a schematic diagram illustrating the spatial scope of
influence in terms of time metric;
FIG. 3 is a schematic block diagram of the apparatus for
determining spatial influences among sections as shown in FIG.
1;
FIG. 4 is a flowchart showing a method for determining spatial
influences among sections;
FIG. 5 is a schematic block diagram of the traffic prediction
apparatus as shown in FIG. 1;
FIG. 6 is a flowchart showing a traffic prediction method; and
FIGS. 7(a) and 7(b) is a schematic diagram illustrating the spatial
relations among sections in a road network according to an
embodiment of the present invention.
DETAILED DESCRIPTION OF PREFERRED EMBODIMENTS
The conventional technologies for traffic prediction fail to take
full consideration of spatial influences among sections, so that
the spatial relations among the sections cannot be fully utilized
in prediction processing. The present invention provides a system
and method for traffic prediction based on space-time relation by
research on how to determine the scope of spatial influence,
allocate weights for spatial influence objects, unify criteria for
evaluation of traffic condition, as well as mine the relation
between the historical traffic conditions of the current road and
the roads within the spatial scope of influence for the current
road. As shown in FIG. 1, a traffic prediction system 1 according
to an embodiment of the present invention mainly comprises: a
section spatial influence determining apparatus 10 for determining,
for each of a plurality of sections for which the traffic condition
is to be predicted, spatial influences on the section by its
neighboring sections; a traffic prediction model establishment
apparatus 20 for establishing, for each of the plurality of
sections, a traffic prediction model by using the spatial
influences determined at the section spatial influence determining
apparatus 10 and historical traffic data of the plurality of
sections; and a traffic prediction apparatus 30 for predicting
traffic condition for each of the plurality of sections for a
future time period by using real-time traffic data and the traffic
prediction model established by the traffic prediction model
establishment apparatus 20. In the traffic prediction system
according to the present embodiment, the apparatus 10, and 30 can
be separated, or any two or all of them can be integrated together.
Also, each of the apparatus 10, 20 and 30 can be formed by separate
or integrated functional units. Additionally, the traffic
prediction system may further comprise: a road network map database
40 for storing road network data; and/or a historical traffic
database 50 for storing historical traffic data for a plurality of
sections. Herein, the historical traffic data may comprise, for a
particular time period in a day, a travel speed at which a vehicle
travels along the section, a travel time period required for a
vehicle to pass through the section and a section congestion
indication. Further, the historical traffic data may be
statistically processed data, such as, for example, historical data
subjected to a conventional statistical process for removing
outliers and peaks or to a difference analysis. The road network
map database 40 may employ a known road network, e.g., a GPS
digital map. The historical traffic database may also be a known
one. Moreover, the real-time traffic data can be obtained from an
existing traffic monitoring system or real-time traffic data
gathering system. Details for known technologies and
functionalities are omitted herein, so as not to obscure the basic
concept of the present invention. The following description focuses
on the section spatial influence determining apparatus 10, the
traffic prediction model establishment apparatus 20 and the traffic
prediction apparatus 30 as mentioned above.
Most of the existing technologies for traffic prediction are based
on part of arterial roads or highways and have thus incomplete road
network and low applicability. The influence of traffic conditions
of some side roads are of significance to urban traffic and may
reflect, directly or indirectly, the traffic condition on
corresponding arterial roads or ring roads. In this regard, the
traffic prediction system of the present invention is developed for
the complete road network.
For purpose of clear description of the concept according to the
present invention, several terminologies will be explained in the
first place.
Section: According to the present invention, a section can be a
link known as a basic road element in most of known road networks,
or a road segment obtained by recombining links in a road network,
or a road segment from one intersection to another in a road
network. As a basic road element, a link is short and has unstable
traffic characteristics. Thus, the section of the present invention
can be a road section reconstructed from links (by integrating, for
example). A section may consist of one or more links, depending on
applications. In this way, the number of prediction objects can be
reduced and the speed and accuracy of prediction can be improved.
Further, a section may be a road segment between key real road
nodes, so that even more useful traffic information can be
available. The section can be set depending on actual
applications.
`Time Order` & `Spatial Order`: A known time sequence model for
space-time statistical data analysis, STARMA model, is assumed for
example. The STARMA model can be generally expressed as:
.times..times..lamda..times..times..PHI..times..times..times..times..time-
s..times..theta..times..times. ##EQU00001## where z.sub.t denotes
an output from a random sequence at time t, p denotes a time
hysteresis order, .lamda..sub.k denotes a spatial hysteresis order,
W.sub.l denotes a spatial operator, which is generally a l-th order
spatial correlation matrix, .phi..sub.kl denotes an auto-regressive
correlation coefficient for a time order of k and a spatial order
of l, a.sub.t, denotes an input into a random sequence at time t,
which is generally a white noise sequence, q denotes an order of
moving average, m.sub.k denotes a moving average spatial hysteresis
order, and .phi..sub.kl denotes a moving average correlation
coefficient for a time order of k and a spatial order of l. Herein,
k is a time order and l is a spatial order, both of which are
hysteresis orders in the above STARMA. For example, when applied to
the prediction processing to predict an output z, from a random
sequence at time t, the respective outputs at t-1, t-2, . . . , t-k
(1.ltoreq.k.ltoreq.p) can be used. In this case, z.sub.t has its
first order output of the random sequence being output z.sub.t-1 at
time t-1 and its k-th order output of the random sequence being
output z.sub.t-k at time t-k. Apparently, the larger p is, the
larger a value of the time order k is, and the larger the time
scope taken into account is. When an output z.sub.t from a random
sequence at time t is to be predicted in STARMA, spatial influences
should be considered in addition to the influences of the
hysteresis sequence outputs at the respective times on z.sub.t.
W.sub.l is a spatial operator representing spatial influence where
l is a spatial order. .lamda..sub.k denotes a scope of spatial
orders to be considered for the time order of k, where
0.ltoreq.l.ltoreq..lamda..sub.k. The l equal to 0 indicates that
only the object to be predicted is considered and the l equal to 1
indicates that the influence on the object to be predicted by its
first order influential object is further considered. In general, a
first order influential object refers to a neighboring object
closest to the object to be predicted in spatial relation and a
second order neighboring object refers to a neighboring object
which is relatively close to the object to be predicted in spatial
relation. In traffic condition prediction, a spatial scope having
influence on the current section can be determined based on
relative spatial location or time metric. For the relative spatial
location, a first order influential object can be a neighboring
section directly adjacent to the current section and a second order
influential object can be a section directly adjacent to a first
order neighboring section of the current section. Likewise, the
first or second order influential object can be determined in terms
of distance, such as at a certain distance from the current
section. For the time metric, on the other hand, a first order
influential object can be a neighboring section of the current
segment in the spatial scope, which is reachable within a
predetermined time period by starting from the current section.
Herein, it is possible to find out a spatial scope reachable from
the current section within a predetermined time period, at an
average travel speed based on historical data of the current
section or at the current speed. The predetermined time period can
be a traffic data gathering period or a multiple thereof, e.g., 5
minutes, half an hour or one hour. FIG. 2 shows a schematic diagram
of a spatial scope of influence for a 5-minute period for example.
The spatial scope reachable from the current section within 5
minutes has a spatial order of 1 and the neighboring sections
within that scope are the first order influential sections of the
current section. Similarly, the spatial scope reachable from the
current section within a period ranging from 5 to 10 minutes has a
spatial order of 2, and the neighboring sections within that scope
are the second order influential sections of the current section.
According to the above equation, for a larger .lamda..sub.k, a
larger value of the spatial order l can be taken and a larger
spatial scope can be taken into account.
Congestion Indication: As noted above, the traffic prediction
system according to the present invention is developed for a
complete road network which includes not only arterial roads, such
as ring roads and highways, but also non-trunk roads such as side
roads. A travel time or a travel speed on a section may be
appropriate for analysis on time sequence of a single section as in
a conventional traffic prediction algorithm, but cannot accurately
reflect the traffic condition in a spatial sense if the spatial
influence relation in the road network is considered. As different
types of roads may have different roadway data and transportation
to capacities, the same travel speed may indicate different levels
of congestion for different classes of roads. A travel speed of 60
km/h, for example, can indicate a certain level of congestion on
highway or a smooth traffic on an ordinary urban street. Thus, it
is desired to consider different physical attributes of roads in
traffic prediction based on space-time relation. As such, a unified
index is required for indicating the levels of congestion of
traffic condition. According to the present invention, Congestion
Indication (CI) is used to indicate a level of congestion for a
section of road, which can refer to a ratio between a real-time
travel time required a vehicle to pass through a section and a
corresponding travel time in a free flow condition:
.function. ##EQU00002## where T.sub.i,t denotes a travel time on a
section X having an index of i at time/time period t and
T.sub.i,normal denotes a travel time in a free flow condition on
the section X having an index of i.
As an alternative, the CI can be a ratio between a real-time travel
speed at which a vehicle travels on a section and a corresponding
travel speed in a free flow condition:
.function. ##EQU00003## Where V.sub.i,t denotes a travel speed on a
section X having an index of i at time/time period t and
V.sub.i,normal denotes a travel speed in a free flow condition on
the section X having an index of i.
In this way, the criterion for measuring traffic condition in a
spatial scope is unified, which can be utilized to measure traffic
in a spatial scope more practically and thereby improve prediction
accuracy. The congestion indication as used herein can be
represented in any other way known to those who skilled in the art
(e.g., the reciprocal of the above CI can be used). Such apparent
variations are encompassed by the scope of the present invention.
During the prediction processing of the present invention, the CI
can be calculated in real-time based on a travel speed or a travel
time gathered by an existing real-time traffic information
gathering system.
With the above explanations of terminologies, the section spatial
influence determining apparatus 10, the traffic prediction model
establishment apparatus 20 and the traffic prediction apparatus 30
will be detailed in the following.
FIG. 3 is a block diagram of the section spatial influence
determining apparatus 10 as shown in FIG. 1. The determination of
section spatial influence is critical for mining spatial relations
among sections according to the present invention, with its major
purpose being to determine influences of the changes in traffic
conditions of neighboring sections of respective orders on the
current section. As shown in FIG. 3, the section spatial influence
determining apparatus 10 according to this embodiment comprises: a
spatial scope determining unit 110 for determining, for each of
sections in a road network, a spatial scope having influence on the
section, wherein the spatial scope is of a spatial order of N,
which is an integer equal to or greater than 1; an influential
section extraction unit 120 for extracting, from the road network,
neighboring sections of the section within the determined spatial
scope, as N-th order influential sections for the section; a
spatial relation determining unit 130 for classifying the spatial
relation between each of the sections and its N-th order
influential sections into one of predefined types of spatial
relation; a correlation learning unit 140 for performing, for the
classified type of spatial relation, correlation analysis on
historical traffic data of the section and its N-th order
influential sections of this type of spatial relation, to learn a
correlation between the section and its N-th order influential
sections for this type of spatial relation; a spatial influence
determining unit 150 for determining spatial influences for the
N-th order influential sections of the section based on the learned
correlation, wherein each of the spatial influences reflects an
extent to which the section is influenced by one of its N-th order
influential sections.
It can be seen from the above explanation for the spatial order
that the spatial scope determining unit 110 can determine the
spatial scope having influence on the current section based on
relative spatial location or time metric, which will be discussed
below respectively.
On one hand, the spatial scope determining unit 110 can determine
the spatial scope having influence on the current section based on
the relative spatial locations between the current section and its
neighboring sections. Herein, a first order influential section can
refer to a neighboring section directly adjacent to the current
section and a second order influential section can refer to a
segment directly adjacent to a first order influential section of
the current section. The spatial adjacency is exemplary only, for
clearly illustrating the present invention, rather than limiting
it. For example, a certain distance can be defined, in which case a
first order influential section may refer to a neighboring section
at the certain distance from the current section, and a second
order influential section may refer to a neighboring section at a
distance twice as long as the certain distance from the current
section, and so on.
On the other hand, the spatial scope determining unit 110 can
determine the spatial scope having influence on the current section
based on the time metric, for example, the spatial scope which is a
range by starting from the current section within a predetermined
time period as shown in FIG. 2. It is possible to find out a
spatial scope by starting from the current section within a
predetermined time period, at an average travel speed based on
historical data of the current segment or at the current speed.
Herein, the predetermined time period can be a traffic data
gathering period or a multiple thereof, so as to facilitate the
gathering and analyzing of traffic data.
The influential section extraction unit 120 can extract, from the
spatial scope of a spatial order of N as determined above, N-th
order influential sections. For the N-th order influential section,
the question is how to take full consideration of their influences
on the current section.
The spatial relation determining unit 130 determines the spatial
relation among a plurality of sections, which is one of the
essential concepts of the present invention. Based on analysis on
road network and historical traffic data, the inventors of the
application realize that different forms of road connectivity have
different levels of influences on traffic conditions of the
connected roads. For a intersection, for example, the traffic
condition of a preceding road has much greater influence than an
intersecting road. Thus, according the present invention, a
plurality of types of spatial relations among sections is
predefined and each of the spatial relations among a plurality of
sections is classified into one of the predefined types of spatial
relations. Table 1 shows the types of spatial relations according
to the embodiment of the present invention as shown in FIG. 7(a),
which are represented with codes 0 and A-H. These nine types of
spatial relations are exemplary common spatial relations only. Any
other spatial relations can be conceived by those who skilled in
the art or any other types of spatial relations can be defined
depending on practical requirements, which are all encompassed by
the scope of the present invention. For example, the types of
spatial relations can also comprise proceed straightforward, left
turn and right turn.
TABLE-US-00001 TABLE 1 Types of spatial relations among segments
Code Description 0 No Relation A Precede Straight B Precede Merge,
C Precede Intersect D Precede Diverge E Succeed Straight F Succeed
Merge G Succeed Intersect H Succeed Diverge
Referring to FIG. 7(a), a plurality of sections in an illustrative
road network is numbered from 1 to 11. In order to clearly explain
the spatial relations among the sections 1-11, the spatial relation
between each section and its first order influential sections is
considered herein for the spatial order of 1, which can be
represented with the spatial relation matrix below:
##EQU00004##
In this matrix, the relation between a section and the section
itself is set to 0, i.e., a section has no relation with itself.
Each row of the matrix represents the spatial relations between the
current section and each of other sections. Herein, only the first
order influential sections are considered. For example, the first
order influential sections of the section 1 are the sections 2 and
3. The spatial relation between the sections 1 and 2 is E, i.e.
precede straight; and the spatial relation between the sections 1
and 3 is G, i.e., precede intersect. The section 1 has no relation
with other sections, which is indicated by 0. In addition, the
spatial relations between each of the sections 2-11 and its first
order influential sections are also indicated in the matrix.
Similarly, an additional spatial relation matrix can be constructed
such that the each row of the matrix represents the spatial
relations between each section and its N-th order influential
sections.
Instead of constructing the above matrix representing the spatial
relations among all the sections, a vector can be constructed, for
each section, to represent the spatial relations between the
segment and its N-th order influential sections for subsequent
storage and calculation. For the segment 5, for example, its first
order influential sections include it directly adjacent sections 2,
4, 6 and 7. Then, the spatial relation vector for the section 5 and
its first order influential sections can be constructed as [B, B,
E, G]. Further, its second order influential sections include
sections directly adjacent to its first order influential sections.
Thus, for each of the sections 2, 4, 6 and 7, it is possible to
determine its first order influential sections first, from which
the second order influential sections of the section 5 can then be
determined. For example, the section 2 has its first order
influential sections being the sections 1 and 5; the section 4 has
its first order influential section being the section 5; the
section 6 has its first influential sections being the segments 5,
8, 9 and 11; and the section 7 has its first influential sections
being the sections 5 and 8. In this case, the second order
influential sections of the section 5 are the sections 1, 8, 9 and
11. The spatial relation vector for the section 5 and its second
order influential sections 1, 8, 9 and 11 can be constructed as [F,
G, H, H]. Obviously, the influential sections can alternatively be
determined in terms of time metric. In this case, for example, the
sections 2, 4, 6 and 7 can be considered as the sections in a
spatial scope with section 5 being the starting point within one
data gathering period at an historical average speed of the section
5.
The approach for determination of spatial relations described
herein is exemplary only. Depending on actual application, those
who skilled in the art can conceive any other types of spatial
relations and any other feasible determination approaches.
As noted above, based on the road network, the spatial relation
determining unit 130 can classify the spatial relation between each
of the sections among a plurality of sections and each of its N-th
order influential sections into one of predefined types of spatial
relation and provide the classified spatial relation to the
correlation learning unit 140.
The correlation learning unit 140 is configured to perform
correlation analysis on historical traffic data of each section and
its N-th order influential sections, to learn a correlation for
each spatial relation as determined. Herein, depending on actual
requirements, short-term, mid-term or long-term historical traffic
data can be utilized. The correlation analysis can be based on a
conventional statistical analysis approach. The above spatial
relations between each of the sections 1-11 and its first order
influential sections are assumed, for example. For the type A,
i.e., downstream proceed, the section pairs involved include the
sections 2 and 1, sections 6 and 5, sections 7 and 8, as well as
sections 10 and 9. In order to learn the section correlation for
the type A, correlation analysis can be performed on the historical
traffic data associated with these section pairs (i.e., sections 2
and 1, sections 6 and 5, sections 7 and 8, as well as sections 10
and 9) by using for example a conventional statistical analysis
approach in which curve charts associated with the historical
traffic data can be plotted with the coordinate axis indicating
time scale. For the sections 2 and 1, for example, the horizontal
and vertical axes are associated with the sections 2 and 1,
respectively, and the time ranges from t-10 to t. In this case, the
traffic data of the sections 2 and 1 at t-10, t-9, . . . , t can be
retrieved from a historical traffic database and the traffic data
points for the respective times can be plotted based on the traffic
data. Herein, the traffic data points can be associated with travel
speed, travel time or congestion indication. Then, a correlation
function can be derived by curve fitting of the respective data
points. As a simple example, for the type A, an approximate
correlation function for the traffic data of the sections 2 and 1
may be a linear function y=ax+b where y denotes the traffic data of
the section 1 at the respective times, x denotes the traffic data
of the section 2 at the respective times and a is denotes the slope
of the linear function. The slope can be used as the correlation
between the sections 2 and 1 for the type A since it can
characterize the linear function. Similarly, correlation analysis
can be performed on the historical traffic data for each of the
other section pairs (i.e., sections 6 and 5, sections 7 and 8, as
well as sections 10 and 9), to obtain a corresponding approximate
correlation function. Then, a value characterizing the correlation
function is used as the correlation of the corresponding section
pair for the type A. In this way, a final correlation can be
obtained as the section correlation for the type A by performing
appropriate statistical processes, such as averaging and median
extracting, on the respective correlations as obtained.
As for the other types of B-H, the above approach is also
applicable for determination of section correlation. Obviously, the
correlation learning unit 140 can apply any other conventional
correlation analysis approaches for the above section correlation
learning. Additionally, the section correlation for each type of
spatial relations can be determined in advance based on historical
traffic data, experiential values or depending on actual
application.
Table 2 gives the results from the section correlation learning for
each type of spatial relations according to this embodiment.
TABLE-US-00002 TABLE 2 Section correlation for each type of spatial
relations Code Description Correlation 0 No Relation 0 A Precede
Straight 1.00 B Precede Merge, 0.80 C Precede Intersect 0.50 D
Precede Diverge 0.50 E Succeed Straight 1.00 F Succeed Merge 0.80 G
Succeed Intersect 0.50 H Succeed Diverge 0.50
The above correlations reflect that the correlation between two
sections having a spatial relation of precede straight or succeed
straight proceed is relatively large, i.e., the level of influence
between them is relatively high. In contrast, the correlation
between two sections having a spatial relation of downstream
intersect, precede diverge, succeed intersect or succeed diverge is
relatively small, i.e., the level of influence between them is
relatively low. It can be seen that the above results are
consistent with the influences among sections in the real world.
That is, at an intersection, a proceeding road imposes much greater
influence on traffic condition than an intersect road.
The correlation learning unit 140 learns the section correlations
for each type of section spatial relations as described above, and
then provides the learned section correlations to the spatial
influence determining unit 150 for determining an extent to which
each section is influenced by each of its N-th order influential
sections.
The spatial influence determining unit 150 is configured to
determine, for each section, levels of spatial influences for the
spatial order of N based on the learned section correlations.
Herein, the spatial relations between each of the above sections
1-11 and its corresponding first order influential sections are
assumed for example. With reference to the above spatial relation
matrix and Table 2, the spatial relation between the sections 1 and
2 is E, i.e., succeed straight; and the spatial relation between
the sections 1 and 3 is G, i.e., succeed intersect. It can be seen
that the section 1 is influenced by the sections 2 and 3. However,
due to different spatial relations, the sections 2 and 3 have
different correlations with the section 1 and thus different levels
of influences on the section 1. The spatial relation between the
sections 1 and 2 is succeed straight and the spatial relation
between the sections 1 and 3 is succeed intersect. Thus, the
section 2 has larger influence on the section 1 when compared with
the section 3. In this embodiment, the spatial influence
determining unit 150 utilizes an influence weight to reflect the
level of influence of each influential section on the current
segment. For example, based on the section correlation between a
particular section and its N-th order influential sections, each of
its N-th order influential sections can be allocated with an
influential weight which can be used to determine the spatial
influence of that influential section on the particular section.
The above section 1 has a correlation of 1.00 with the section 2
and a correlation of 0.50 with the section 3, in which case the
influence weight allocated to the section 2 can be calculated as
1.00/(1.00+0.50)=0.67 and the influence weight allocated to the
section 3 as 0.50/(1.00+0.50)=0.33. Similarly, the section 2 has a
correlation of 1.00 with the section 1 and a correlation of 0.80
with the section 5, in which case the influence weight allocated to
the section 1 can be calculated as 1.00/(1.00+0.80)=0.55 and the
influence weight allocated to the section 5 as
0.80/(1.00+0.80)=0.45. For each section as current section, the sum
of influence weights allocated to all of its first order
influential sections can be set as 1 and each of the influence
weights can be used as the spatial influence on the current section
by one of its N-th order influential sections. In this way, it is
possible to easily and effectively reflect the levels of influences
on the current section by its influential sections, thereby
simplifying corresponding calculation. However, this is exemplary
only; other influence weight levels or ratios can be employed
alternatively depending on actual situation and application. The
spatial influence determining unit 150 is further configured to
determine the spatial influence on the current section by each of
its first order influential sections. As an example, the spatial
influences can be directly represented using the values of the
respective influence weights, to obtain the following spatial
influence matrix W.sub.1.
##EQU00005##
As an alternative, spatial influence vectors for the respective
sections can be obtained for the spatial order of 1. For example,
the spatial influence vector for the section 5 can be [0.26, 0.26,
0.32, 0.16].
As noted above, the section spatial influence determining apparatus
10 can determine, for a spatial order of N, levels of influences on
each section by its N-th order neighboring sections, such that the
influences on the current section by changes in traffic conditions
of the neighboring sections can be introduced into the prediction
process. As such, in actual prediction, any change in traffic
condition at a node or on a section can be rapidly reflected in the
corresponding spatial scope, which is impossible for the prediction
algorithm considering only one single segment.
Furthermore, with the above determination approach for the spatial
order of 1, the section spatial influence determining apparatus 10
can determine, for a number of other spatial orders, levels of
influences on each section by its neighboring sections. In other
words, for a changed spatial order N, spatial influences for the
changed spatial order N can be determined for each section by the
spatial scope determining step, the influential section extraction
step, the spatial relation determining unit, the correlation
learning unit and the spatial influence determining unit.
In this way, the spatial relations among all the sections in the
road traffic network can be fully utilized to obtain, for each of a
plurality of different spatial orders, influence of the changes in
traffic condition of neighboring sections on the current section,
such that the overall traffic condition can be reflected.
Additionally, in actual prediction, it is possible to select a
suitable spatial order based on the time period or traffic
condition to be predicted, so as to determine a spatial scope to be
considered for prediction. As such, the traffic prediction can be
more flexible and effective.
In addition, the section spatial influence determining apparatus 10
can further comprise a storage unit (not shown) for storing, for
each section, the determined spatial influences for at least one
spatial order N, for example in a matrix or vector form as
described above.
FIG. 4 is a flowchart showing the method for determining spatial
influences among sections. In the section spatial influence
determining process performed by the section spatial influence
determining apparatus 10, for each of sections in a road network, a
spatial scope having influence on the section is determined at step
400, as shown in FIG. 4. At step 402, neighboring sections of the
section within the determined spatial scope are extracted from the
road network, as N-th order influential sections for the segment.
At step 404, the spatial relation between each of the sections and
each of its N-th order influential sections is classified into one
of predefined types of spatial relation. At step 406, for the
classified type of spatial relation, correlation analysis is
performed on historical traffic data of the section and its N-th
order influential sections of this type of spatial relation, to
learn a correlation between the section and its N-th order
influential sections for this type of spatial relation. At step
408, spatial influences of spatial order N for the section are
determined based on the learned section correlation. Herein, the
spatial order N can be changed and the spatial influences of a
plurality of spatial orders can be determined for each section by
repeating steps 400-408. At step 410, the determined spatial
influences for at least one spatial order N is stored for each
section.
The foregoing describes in detail the section spatial influence
determining apparatus 10 in the traffic prediction system 1
according to the present invention, as well as the method for
determining spatial influences among section as performed by the
apparatus 10. The apparatus 10 and its corresponding method are
capable of determining, for a plurality of spatial orders, levels
of influences on each section by its neighboring sections. The
determined spatial influences can be used as a spatial factor in
traffic prediction model establishment and traffic prediction. In
this way, the spatial relations among sections in a road traffic
network itself can be fully utilized and the influence of changes
in traffic conditions of neighboring sections on the current
segment is considered for various spatial orders.
A detailed description of the traffic prediction model
establishment section in the traffic prediction system 1 of the
present invention will be given below. The traffic prediction
system 1 comprises a traffic prediction model establishment
apparatus 20 which is configured to establish, for each of the
plurality of sections to be predicted, a traffic prediction model
by using the spatial influences determined at the section spatial
influence determining apparatus 10 and historical traffic data of
the plurality of sections. As an example, the traffic prediction
model establishment apparatus 20 may obtain historical traffic data
of a plurality of sections for a particular period, estimate
individual parameters for a predetermined prediction model based on
the obtained historical traffic data and the spatial influence for
each of the plurality of sections as determined at the section
spatial influence determining apparatus 10, and substitute, for
each section, the estimated parameters and the spatial influences
for the section into the predetermined prediction model, so as to
establish a traffic prediction model of the section for the
particular period. Also, based on the obtained historical traffic
data and the spatial influence for each of the plurality of
sections, the traffic prediction model establishment apparatus 20
can multiply the spatial influence for each of the plurality of
sections with the historical traffic data of all the neighboring
sections of the section for the spatial order associated with the
spatial influence, so as to obtain a sample for model
establishment. In this case, the estimation of parameters is
conducted based on the obtained sample. However, the generation of
the sample is optional and the parameters can be estimated by
directly inputting the historical traffic data and the spatial
influences.
In the traffic prediction according to the present invention, a
time sequence model, which is commonly used in statistical analysis
and incorporates spatial relations, can be utilized, including a
Space-Time Auto Regression (STAR) model and a Space-Time Auto
Regression Moving Average (STARMA) model, both of which are
suitable for analysis on space-time statistical data.
Alternatively, any other suitable time sequence model incorporating
spatial relations can be used. The spatial influences among
sections as determined by the section spatial influence determining
apparatus 10 can be used as a spatial operator in a prediction
model, such that the influences of neighboring sections on the
current section to be predicted can be taken into account during
model establishment.
In this embodiment, the historical traffic data of a plurality of
sections for a number of periods can be retrieved from a historical
traffic database. For each section, the traffic prediction model
establishment apparatus 20 estimates, for a time order and a
spatial order specified for the employed time sequence model,
parameters for the predetermined time sequence model based on the
historical data and spatial influences for the specified time and
spatial orders. Additionally, the traffic prediction model
establishment apparatus 20 is configured to substitute, for each
section, the estimated parameters and the spatial influences for
the section into the predetermined time sequence model, so as to
establish a traffic prediction model of the section for the
specified time and spatial orders. Herein, the traffic prediction
model establishment apparatus 20 may utilize a conventional
modeling approach. In the following, the sections in the schematic
diagram of the road network as shown in FIG. 7(b) are taken as an
example to describe the modeling process by the traffic prediction
model establishment apparatus 20 based on the STAR model.
The STAR model can be represented as follows:
.times..times..lamda..times..times..PHI..times..times. ##EQU00006##
where z.sub.t denotes an output from a random sequence at time t, p
denotes a delayed time order, .lamda..sub.k denotes a delayed
spatial order, W.sub.l denotes a spatial operator of the STAR
model, which is represented as a l-th order spatial influence
vector or matrix as determined by the section spatial influence
determining apparatus 10 of the present invention, and .phi..sub.kl
denotes a coefficient for a time order of k and a spatial order of
l, i.e., the coefficient to be estimated. When compared with the
above STARMA model, the items of moving average and white noise
sequence are omitted in the STAR model as both items are mainly
used for model adjustment and are not essential for construction of
the model. Therefore, the STAR model is adopted herein to clearly
illustrate the basic concept of the present invention.
When applied to traffic prediction, z.sub.t represents traffic
condition of the section to be predicted, i.e., traffic data for a
period centered at time t, which reflects the traffic condition,
such as congestion level, for the period. The traffic data can be a
travel speed, a travel time or a congestion indication. The
prediction is based on the historical traffic data, that is, on the
traffic data for time periods centered at respective times t-1,
t-2, . . . , t-k. W.sub.l denotes a l-th order spatial influence
vector or matrix. Then, the establishment of the model mainly
involves estimation of the efficient .phi..sub.kl for a time order
of k and a spatial order of l.
As an example, the estimation of parameters is performed under an
assumption that p=2 and .lamda..sub.k=2. In this case, the
prediction model becomes the following equation (1):
z.sub.t=.phi..sub.11.times.S.sub.1,t-1+.phi..sub.12.times.S.sub.2,t-1+.ph-
i..sub.21.times.S.sub.1,t-2+.phi..sub.22.times.S.sub.2,t-2,S=W.sub.lz.sub.-
t. (1)
Herein, the parameters to be predicted include .phi..sub.11,
.phi..sub.12, .phi..sub.21 and .phi..sub.22. The section 1 as shown
in FIG. 7(b) is assumed as the current section, which has 6 first
order neighboring sections numbered as 2 to 7, respectively, and 15
second order neighboring sections numbered as 8 to 22,
respectively. The section spatial influence determining apparatus
10 is used to determine the spatial relations between the section 1
and its first order and second order neighboring sections. The
types of spatial relations used herein include proceed
straightforward, left turn and right turn. Further, the section
spatial influence determining apparatus 10 is used to learn, for
the spatial orders of 1 and 2, the section correlations for these
types of spatial relations (i.e., proceed straightforward, left
turn and right turn) as follows: proceed straightforward: 1; left
turn: 0.8 and right turn: 0.6.
Next, according to the above method, the section spatial influence
determining apparatus 10 obtains the spatial influences on the
section 1 by its first order and second order neighboring sections,
as follows:
.times..times..times..times..times..times..times..times..times..times..ti-
mes..times..times..times. ##EQU00007##
.times..times..times..times..times..times..times..times..times..times.
##EQU00007.2##
.times..times..times..times..times..times..times..times..times..times..ti-
mes..times..times. ##EQU00007.3##
.times..times..times..times..times..times..times..times..times..times..ti-
mes..times..times..times..times..times..times..times..times..times..times.-
.times..times..times..times..times..times..times..times..times..times.
##EQU00007.4##
Again, the spatial influences can be determined by the section
spatial influence determining apparatus 10 in advance.
The historical traffic data can be retrieved from the historical
traffic database. Herein, the congestion indications are used as
the traffic data, which are exemplified as follows:
TABLE-US-00003 TABLE 3 Historical traffic data for the spatial
order of 1 Section Time Period 1 2 3 4 5 6 7 2009_7_1_10 1.472
1.366 1.365 1.097 1.489 1.309 1.921 2009_7_1_11 1.913 1.298 1.267
1.469 1.654 1.722 1.921 . . . . . . . . . . . . . . . . . . . . . .
. . 2009_m_n_t 1.398 1.093 1.170 1.386 1.406 1.446 1.743
TABLE-US-00004 TABLE 4 Historical traffic data for the spatial
order of 2 Section Time Period 1 8 9 10 11 12 13 14 15 16 17 18 19
20 21 22 2009_7_1_10 1.472 1.292 1.911 1.721 . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . 2009_7_1_11 1.913
1.424 1.656 1.232 . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . 2009_m_n_t
1.398 1.292 1.258 1.265 . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . .
Each row of Table 3 or Table 4 constitutes a traffic data vector
z.sub.t for a corresponding time period.
With the above spatial influences W.sub.1 and W.sub.2 as well as
the historical traffic data vectors, a sample generation unit 240
can calculate S.sub.l,t=W.sub.l.times.z.sub.t as a model sample for
estimation of parameters. Particularly, with respect to z.sub.t for
the period centered at time t, S.sub.1,t-1=W.sub.1.times.z.sub.t-1,
S.sub.2,t-1=W.sub.2.times.z.sub.t-1,
S.sub.1,t-2=W.sub.1.times.z.sub.t-2 and
S.sub.2,t-2=W.sub.2.times.z.sub.t-2. The respective values of the
parameters .phi..sub.11, .phi..sub.12, .phi..sub.21 and
.phi..sub.22 can be calculated by substituting z.sub.t and
S.sub.1,t-1, . . . , S.sub.2,t-2 for each period into equation (1).
Herein, depending on actual requirements, the estimation of the
parameters can be based on short-term, mid-term or long-term
historical traffic data. A number of sets of estimated parameter
values can be obtained, from which optimal estimated parameter
values can be found by using a conventional statistical evaluation
approach, e.g., by analyzing statistical values such as standard
deviation and variance.
For the section 1, the traffic prediction model for the time order
of 2 and the spatial order of 2 can be established by substituting
the estimated parameters and the spatial influences for the section
1 into equation (1), as follows:
z.sub.t=0.17499.times.W.sub.1z.sub.t-1+0.37183.times.W.sub.2z.sub.t-1+0.1-
3391.times.W.sub.1z.sub.t-2+0.23458.times.W.sub.2z.sub.t-2. (2)
The specific calculation processes for the above parameter
estimation and statistical evaluation can be based on conventional
approaches and the details thereof can be omitted herein.
Furthermore, for a changed time period and/or time order and/or
spatial order, the traffic prediction model establishment apparatus
20 can establish, for each section, a corresponding traffic
prediction model based on the historical traffic data and spatial
influences corresponding to the changed time period and/or time
order and/or spatial order. In the above example, a traffic
prediction model is established for the section 1 with respect to
the time period centered at time t, the spatial order of 2 and the
time order of 2. In addition, the traffic prediction model
establishment apparatus 20 can establish for the section 1 a
traffic prediction model with respect to a time period centered at
time t, a spatial order of 3 and a time order of 2, a traffic
prediction model with respect to a time period centered at time
t+1, a spatial order of 3 and a time order of 3, and the like. In
this way, a section may have a number of traffic prediction models
each corresponding to one of different time periods and/or time
orders and/or spatial orders. As such, prediction models for
different time scopes and spatial scopes can be established for
each section by incorporating different situations of the section
for different time periods, such that the traffic prediction can be
more flexible and effective. Also, the traffic prediction model
establishment apparatus 20 can be configured to store at least one
traffic prediction model established for each section. A stored
prediction model for a corresponding section can be selected for
traffic prediction.
Details of the traffic prediction section in the traffic prediction
system 1 according to the present invention will be given below.
The traffic prediction system 1 comprises a traffic prediction
apparatus 30 adapted for selecting from the models established by
the traffic prediction model establishment apparatus 20 a
prediction model for each section and performing traffic prediction
for a future time period based on real-time traffic data. FIG. 5 is
a structural diagram of the traffic prediction apparatus 30 as
shown in FIG. 1, which comprises: a prediction input obtaining unit
310 for obtaining real-time traffic data for a plurality of
sections within one or more time periods, as a prediction input; a
traffic prediction model selection unit 320 for selecting a traffic
prediction model for each of the sections whose traffic is to be
predicted, based on a future time period for which the prediction
is to be made and/or a specified time order and/or spatial order,
wherein the traffic prediction model is a time sequence model
incorporating is spatial relation, and the spatial relation is
represented by spatial influences among the sections as determined
by the section spatial influence determining apparatus 10 (e.g.,
the traffic prediction model can be established by the traffic
prediction model establishment apparatus 20 according to the above
procedures); and a traffic prediction unit 330 for predicting
traffic of each of the section for a future time period after a
specified time period by using the prediction input and the
selected traffic prediction model. The traffic prediction apparatus
30 can further comprise: a data difference analysis unit 340 for
analyzing the difference between the real-time traffic data
obtained by the prediction input obtaining unit 310 and the
historical traffic data, adjusting the obtained real-time traffic
data based on the analysis result, and using the adjusted real-time
traffic data as the prediction input. The data difference analysis
unit 340 can be configured to adjust the real-time traffic data
using a conventional statistical averaging approach, so as to
remove outliers and peaks from the real-time traffic data and to
improve the accuracy of the prediction input. The prediction input
obtaining unit 310 is configured for obtaining from an existing
real-time traffic monitoring system the real-time traffic data for
a plurality of sections, including a travel speed or travel time,
and for calculating in real-time a congestion indication based on
the travel speed or travel time. The traffic prediction model
selection unit 320 is configured for selecting, for each of the
sections whose traffic is to be predicted, from the traffic
prediction models established by the traffic prediction model
establishment apparatus 20 traffic prediction models for different
time orders and/or spatial orders, based on a future time period to
be predicted. As a simple example, this selection can be specified
by an operator. For an arterial road in rush hours, for example, a
prediction model having a large time order and a large spatial
order can be selected, so as to consider influences in a large time
and spatial scope. For a side road in non-rush hours, in contrast,
a prediction model with a small time order and a small spatial
order can be selected. In addition, for prediction models
established from short-term, mid-term and long-term historical
traffic data, the traffic prediction model can be selected
depending on whether a short-term, mid-term or long-term traffic is
to be predicted. The traffic prediction unit 330 is configured for
predicting traffic for a future time period based on the prediction
input from the prediction input obtaining unit 310 or the data
difference analysis unit 340 and the selected traffic prediction
model. As for the above example, in order to predict the traffic
z.sub.t of the section 1 for a time period centered at time t, the
prediction inputs can be obtained based on the real-time traffic
data z.sub.t-1 and z.sub.t-2: S.sub.1,t-1=W.sub.1.times.z.sub.t-1,
S.sub.2,t-1=W.sub.2.times.z.sub.t-1,
S.sub.1,t-2=W.sub.1.times.z.sub.t-2 and
S.sub.2,t-2=W.sub.2.times.z.sub.t-2. These prediction inputs are
then substituted into equation (2) for calculating z.sub.t as the
prediction result.
The traffic prediction apparatus 30 can further comprise a
prediction result output unit (not shown) for storing and
outputting the prediction result.
FIG. 6 is a flowchart of the traffic prediction method, which
illustrates the operation of the traffic prediction apparatus 30.
At step 600, the prediction input obtaining unit obtains real-time
traffic data for a plurality of sections within one or more time
periods. At step 602, the data difference analysis unit 340
analyzes the difference between the real-time traffic data obtained
by the prediction input obtaining unit 310 and the historical
traffic data, adjusts the obtained real-time traffic data based on
the analysis result, and using the adjusted real-time traffic data
as the prediction input. At step 604, the traffic prediction model
selection unit 320 selects a traffic prediction model for each of
the sections whose traffic is to be predicted, based on a future
time period for which the prediction is to be made. At step 606,
the traffic prediction unit 330 predicts traffic of each of the
section for a future time period after a specified time period by
using the prediction input and the selected traffic prediction
model. At step 608, the prediction result output unit stores and
outputs the prediction result.
The traffic prediction system of the present invention has been
described above, which is capable of predicting future traffic and
calculating compensation for current traffic, in order to increase
traffic coverage rate. For example, for the sections as shown in
FIG. 7(a), the traffic condition of the section 5 can be estimated
given the predicted traffic conditions for the sections 2 and 4. In
the case where the sections and 4 each have a high level of
congestion, the traffic on the section 5 can be considered to be
congested.
It should be noted that the foregoing illustrates the solutions of
the present invention by way of example only and is not intended to
limit the present invention to the steps and element structures as
described above. It is possible to adjust and modify such steps and
element structures as desired. Thus, some of the steps and elements
are not essential for implementing the general concept of the
present invention. Accordingly, the essential technical features of
the present invention are limited by only the minimum requirements
for implementing the general concept of the present invention,
rather than the above particular embodiments.
To this end, the present invention has been disclosed with
reference to the preferred embodiments thereof. It can be
appreciated that any other modifications, alternatives and
additions can be made by those who skilled in the art without
departing from the spirits and scope of the present invention.
Therefore, the scope of the present invention is not limited to the
above particular embodiments, but only limited by the claims as
attached.
* * * * *